Visualize Any AI Model File Instantly with Netron

// Before you start

What you need

A model file to inspect (ONNX, PyTorch, TensorFlow, Keras, Safetensors, GGUF, etc.)
A modern browser (Chrome, Firefox, Edge) — OR —
Windows, macOS, or Linux desktop for the installed version
Optional: Python 3.x installed if you want the pip version

Step 1 of 6

Choose your install method

1 min

Netron has three ways to run: browser (nothing to install), desktop app (best experience), or Python package. Pick the one that fits you. If you just want to try it right now, the browser version works immediately. If you plan to use it regularly, install the desktop app.

Terminal · mac

$ # OPTION A — Browser (no install needed):

$ # Visit https://netron.app in your browser

$ # OPTION B — Windows desktop:

$ winget install -s winget netron

$ # OPTION C — macOS desktop:

$ brew install --cask netron

$ # OPTION D — Python:

$ pip install netron

What you should see

Browser: netron.app loads in your tab. Desktop: Netron appears in your applications. Python: 'Successfully installed netron-x.x.x'

This might happen

winget or brew command not found

For Windows, update Windows first (winget ships with modern Windows 10/11). For macOS, install Homebrew from brew.sh first. Or just use the browser version at netron.app — no install needed.

Step 2 of 6

Get a sample model file to open

2 min

You need a model file to visualize. If you already have one (a .onnx, .pt, .tflite, .h5, .safetensors, or similar file), skip this step. Otherwise, download one of Netron's official sample files to practice with. The SqueezeNet ONNX file is a good starting point — it's small and well-structured.

Terminal · mac

$ # Download the SqueezeNet ONNX sample model:

$ curl -L -o squeezenet.onnx https://media.githubusercontent.com/media/onnx/models/main/validated/vision/classification/squeezenet/model/squeezenet1.0-3.onnx

$ # OR just open it directly in the browser version:

$ # Visit https://netron.app and click the link for 'squeezenet' on the page

What you should see

A file called squeezenet.onnx appears in your current folder (about 5 MB).

This might happen

curl not found on Windows

Open the URL directly in your browser to download the file, or use the browser version of Netron which can open sample files with one click.

Step 3 of 6

Open your model in the browser version

1 min

If you chose the browser route, this is all you do. Go to netron.app, then drag your model file onto the page — or click the folder icon to browse for it. Netron reads the file entirely in your browser; nothing is uploaded to any server.

Terminal · mac

$ # No command needed.

$ # 1. Go to https://netron.app

$ # 2. Drag your .onnx (or other model) file onto the browser window

$ # OR click the open-folder icon and select your file.

What you should see

A graph diagram appears showing all the layers and connections in your model. You can scroll, zoom, and click any node to see its properties.

This might happen

The page loads but nothing happens after dropping the file

Make sure the file extension is supported (.onnx, .pt, .tflite, .h5, .safetensors, .gguf, etc.). Very large files (>500 MB) may take 30–60 seconds to render — wait before assuming it failed.

Step 4 of 6

Open your model in the desktop app

1 min

If you installed the desktop app, simply launch Netron from your applications list. It opens with a file picker. Select your model file. The desktop version handles larger files more reliably than the browser and remembers recently opened files.

Terminal · mac

$ # No command needed.

$ # Launch Netron from your Start Menu (Windows),

$ # Applications folder (macOS), or app launcher (Linux).

$ # Then: File > Open > select your model file.

$ # OR double-click a supported model file — Netron registers itself

$ # as the default viewer for common model formats.

What you should see

Netron opens and renders the model graph. The window title shows your filename.

This might happen

Double-clicking a .onnx file opens something else instead of Netron

Right-click the file, choose 'Open With', and select Netron. You can set it as the default from there.

Step 5 of 6

Open your model via Python (optional)

2 min

If you installed Netron via pip, you can launch it from the command line pointing at any model file. This is handy if you are already working in a Python environment and want to inspect a model you just trained or downloaded.

Terminal · mac

$ # Replace 'yourmodel.onnx' with your actual filename:

$ netron yourmodel.onnx

$ # Netron will print a local URL and open your browser automatically.

$ # Example output:

$ # Serving 'yourmodel.onnx' at http://localhost:8080

What you should see

Your browser opens automatically to http://localhost:8080 and displays the model graph.

This might happen

'netron' command not found after pip install

Your Python scripts folder may not be on your PATH. Try running 'python -m netron yourmodel.onnx' instead.

Step 6 of 6

Explore the model graph

5 min

Once your model is open, you can interact with the graph. Click any node (a box in the diagram) to see its operator type, input/output shapes, and attributes in a side panel. Use the scroll wheel or pinch gesture to zoom. Use the search icon (magnifying glass) to find a specific layer by name. The Properties panel on the right shows the full model metadata when you click the background.

Terminal · mac

$ # No commands — this is all mouse/trackpad interaction.

$ # Zoom: scroll wheel or pinch

$ # Pan: click and drag the background

$ # Inspect a layer: click any node box

$ # Search: click the magnifying glass icon (top toolbar)

$ # Export graph as image: File > Export as PNG (desktop) or

$ # click the export icon (browser)

What you should see

Clicking a node shows its name, operator type, input/output tensor shapes, and any hyperparameters (kernel size, stride, etc.) in the side panel.

// Status

cooked. baked. worked.

You can open any supported AI model file and see a full interactive graph of its layers, operators, and tensor shapes — useful for understanding model architecture, debugging, or documentation.

// the honest bit

The honest part

Netron is a viewer only — you cannot edit, retrain, or convert models with it. It shows architecture and metadata but does not display actual trained weight values in a human-readable way (weights are stored as raw tensors). Some newer or exotic model formats may render partially or not at all; check the README for the current support list. Very large models (multi-GB) can be slow or crash the browser version — use the desktop app for those.